在用户给定场景中,为了利用估计出的半透明材质参数渲染得到与输入图片具有相似外观的半透明物体,提出一种基于颜色直方图匹配对单张图片进行半透明材质估计的算法.对于均匀半透明材质,首先将材质估计问题转化为一个能量最小化的过程,能量函数被定义为输入图与渲染图之间颜色直方图的差别,并通过共轭梯度法优化能量函数,得到材质估计结果;对于非均匀半透明材质,通过材质聚类将所得材质视为均匀半透明材质分别进行估计.实验结果表明,由于避免了形状恢复以及光照条件估计引进误差,该算法能够高质量地进行复杂几何形体、复杂光照条件下的半透明材质估计.
In this paper, we present a translucent material estimation method based on color histogram matching. Our algorithm could estimate translucent material parameters from a single input image, producing similar translucent appearances in the synthetic image to the input image. The algorithm treats the translucent material estimation for homogeneous material as an energy minimization problem, which was defined as the difference between the color histograms of the input image and the synthetic image. Our method uses conjugate gradient algorithm to minimize energy function and get the translucent material estimation results. Heterogeneous translucent material can also be estimated after material classification. As the results show, our method could produce convincing results under complicated 3D shape and lighting conditions, avoiding the inaccuracy introduced by shape and lighting estimation.